It doesn’t make sense in my case, though I noted that works as well. I’d prefer to use the OffsetArrays (which allow for the 0 or other index-based indexing) because of the way it interacts with the convolution function. There are definitely other ways to implement what I’d like to do, and I will probably have to go with one of those, but LinearMaps are simply more elegant and I would like to understand why this doesn’t work so I can use them as often as possible.
Also, I could understand there being some issues between the OffsetArray indexing and LinearMaps (there’s no documentation I can find on whether LinearMaps supports OffsetArrays or not), but I’m surprised there seems to be a difference between the forward and transpose functionality.